xref: /linux/drivers/ntb/hw/amd/ntb_hw_amd.c (revision a8fe58cec351c25e09c393bf46117c0c47b5a17c)
1 /*
2  * This file is provided under a dual BSD/GPLv2 license.  When using or
3  *   redistributing this file, you may do so under either license.
4  *
5  *   GPL LICENSE SUMMARY
6  *
7  *   Copyright (C) 2016 Advanced Micro Devices, Inc. All Rights Reserved.
8  *
9  *   This program is free software; you can redistribute it and/or modify
10  *   it under the terms of version 2 of the GNU General Public License as
11  *   published by the Free Software Foundation.
12  *
13  *   BSD LICENSE
14  *
15  *   Copyright (C) 2016 Advanced Micro Devices, Inc. All Rights Reserved.
16  *
17  *   Redistribution and use in source and binary forms, with or without
18  *   modification, are permitted provided that the following conditions
19  *   are met:
20  *
21  *     * Redistributions of source code must retain the above copyright
22  *       notice, this list of conditions and the following disclaimer.
23  *     * Redistributions in binary form must reproduce the above copy
24  *       notice, this list of conditions and the following disclaimer in
25  *       the documentation and/or other materials provided with the
26  *       distribution.
27  *     * Neither the name of AMD Corporation nor the names of its
28  *       contributors may be used to endorse or promote products derived
29  *       from this software without specific prior written permission.
30  *
31  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
32  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
33  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
34  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
35  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
36  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
37  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
38  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
39  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
40  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
41  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
42  *
43  * AMD PCIe NTB Linux driver
44  *
45  * Contact Information:
46  * Xiangliang Yu <Xiangliang.Yu@amd.com>
47  */
48 
49 #include <linux/debugfs.h>
50 #include <linux/delay.h>
51 #include <linux/init.h>
52 #include <linux/interrupt.h>
53 #include <linux/module.h>
54 #include <linux/acpi.h>
55 #include <linux/pci.h>
56 #include <linux/random.h>
57 #include <linux/slab.h>
58 #include <linux/ntb.h>
59 
60 #include "ntb_hw_amd.h"
61 
62 #define NTB_NAME	"ntb_hw_amd"
63 #define NTB_DESC	"AMD(R) PCI-E Non-Transparent Bridge Driver"
64 #define NTB_VER		"1.0"
65 
66 MODULE_DESCRIPTION(NTB_DESC);
67 MODULE_VERSION(NTB_VER);
68 MODULE_LICENSE("Dual BSD/GPL");
69 MODULE_AUTHOR("AMD Inc.");
70 
71 static const struct file_operations amd_ntb_debugfs_info;
72 static struct dentry *debugfs_dir;
73 
74 static int ndev_mw_to_bar(struct amd_ntb_dev *ndev, int idx)
75 {
76 	if (idx < 0 || idx > ndev->mw_count)
77 		return -EINVAL;
78 
79 	return 1 << idx;
80 }
81 
82 static int amd_ntb_mw_count(struct ntb_dev *ntb)
83 {
84 	return ntb_ndev(ntb)->mw_count;
85 }
86 
87 static int amd_ntb_mw_get_range(struct ntb_dev *ntb, int idx,
88 				phys_addr_t *base,
89 				resource_size_t *size,
90 				resource_size_t *align,
91 				resource_size_t *align_size)
92 {
93 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
94 	int bar;
95 
96 	bar = ndev_mw_to_bar(ndev, idx);
97 	if (bar < 0)
98 		return bar;
99 
100 	if (base)
101 		*base = pci_resource_start(ndev->ntb.pdev, bar);
102 
103 	if (size)
104 		*size = pci_resource_len(ndev->ntb.pdev, bar);
105 
106 	if (align)
107 		*align = SZ_4K;
108 
109 	if (align_size)
110 		*align_size = 1;
111 
112 	return 0;
113 }
114 
115 static int amd_ntb_mw_set_trans(struct ntb_dev *ntb, int idx,
116 				dma_addr_t addr, resource_size_t size)
117 {
118 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
119 	unsigned long xlat_reg, limit_reg = 0;
120 	resource_size_t mw_size;
121 	void __iomem *mmio, *peer_mmio;
122 	u64 base_addr, limit, reg_val;
123 	int bar;
124 
125 	bar = ndev_mw_to_bar(ndev, idx);
126 	if (bar < 0)
127 		return bar;
128 
129 	mw_size = pci_resource_len(ndev->ntb.pdev, bar);
130 
131 	/* make sure the range fits in the usable mw size */
132 	if (size > mw_size)
133 		return -EINVAL;
134 
135 	mmio = ndev->self_mmio;
136 	peer_mmio = ndev->peer_mmio;
137 
138 	base_addr = pci_resource_start(ndev->ntb.pdev, bar);
139 
140 	if (bar != 1) {
141 		xlat_reg = AMD_BAR23XLAT_OFFSET + ((bar - 2) << 3);
142 		limit_reg = AMD_BAR23LMT_OFFSET + ((bar - 2) << 3);
143 
144 		/* Set the limit if supported */
145 		limit = base_addr + size;
146 
147 		/* set and verify setting the translation address */
148 		write64(addr, peer_mmio + xlat_reg);
149 		reg_val = read64(peer_mmio + xlat_reg);
150 		if (reg_val != addr) {
151 			write64(0, peer_mmio + xlat_reg);
152 			return -EIO;
153 		}
154 
155 		/* set and verify setting the limit */
156 		write64(limit, mmio + limit_reg);
157 		reg_val = read64(mmio + limit_reg);
158 		if (reg_val != limit) {
159 			write64(base_addr, mmio + limit_reg);
160 			write64(0, peer_mmio + xlat_reg);
161 			return -EIO;
162 		}
163 	} else {
164 		xlat_reg = AMD_BAR1XLAT_OFFSET;
165 		limit_reg = AMD_BAR1LMT_OFFSET;
166 
167 		/* split bar addr range must all be 32 bit */
168 		if (addr & (~0ull << 32))
169 			return -EINVAL;
170 		if ((addr + size) & (~0ull << 32))
171 			return -EINVAL;
172 
173 		/* Set the limit if supported */
174 		limit = base_addr + size;
175 
176 		/* set and verify setting the translation address */
177 		write64(addr, peer_mmio + xlat_reg);
178 		reg_val = read64(peer_mmio + xlat_reg);
179 		if (reg_val != addr) {
180 			write64(0, peer_mmio + xlat_reg);
181 			return -EIO;
182 		}
183 
184 		/* set and verify setting the limit */
185 		writel(limit, mmio + limit_reg);
186 		reg_val = readl(mmio + limit_reg);
187 		if (reg_val != limit) {
188 			writel(base_addr, mmio + limit_reg);
189 			writel(0, peer_mmio + xlat_reg);
190 			return -EIO;
191 		}
192 	}
193 
194 	return 0;
195 }
196 
197 static int amd_link_is_up(struct amd_ntb_dev *ndev)
198 {
199 	if (!ndev->peer_sta)
200 		return NTB_LNK_STA_ACTIVE(ndev->cntl_sta);
201 
202 	/* If peer_sta is reset or D0 event, the ISR has
203 	 * started a timer to check link status of hardware.
204 	 * So here just clear status bit. And if peer_sta is
205 	 * D3 or PME_TO, D0/reset event will be happened when
206 	 * system wakeup/poweron, so do nothing here.
207 	 */
208 	if (ndev->peer_sta & AMD_PEER_RESET_EVENT)
209 		ndev->peer_sta &= ~AMD_PEER_RESET_EVENT;
210 	else if (ndev->peer_sta & AMD_PEER_D0_EVENT)
211 		ndev->peer_sta = 0;
212 
213 	return 0;
214 }
215 
216 static int amd_ntb_link_is_up(struct ntb_dev *ntb,
217 			      enum ntb_speed *speed,
218 			      enum ntb_width *width)
219 {
220 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
221 	int ret = 0;
222 
223 	if (amd_link_is_up(ndev)) {
224 		if (speed)
225 			*speed = NTB_LNK_STA_SPEED(ndev->lnk_sta);
226 		if (width)
227 			*width = NTB_LNK_STA_WIDTH(ndev->lnk_sta);
228 
229 		dev_dbg(ndev_dev(ndev), "link is up.\n");
230 
231 		ret = 1;
232 	} else {
233 		if (speed)
234 			*speed = NTB_SPEED_NONE;
235 		if (width)
236 			*width = NTB_WIDTH_NONE;
237 
238 		dev_dbg(ndev_dev(ndev), "link is down.\n");
239 	}
240 
241 	return ret;
242 }
243 
244 static int amd_ntb_link_enable(struct ntb_dev *ntb,
245 			       enum ntb_speed max_speed,
246 			       enum ntb_width max_width)
247 {
248 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
249 	void __iomem *mmio = ndev->self_mmio;
250 	u32 ntb_ctl;
251 
252 	/* Enable event interrupt */
253 	ndev->int_mask &= ~AMD_EVENT_INTMASK;
254 	writel(ndev->int_mask, mmio + AMD_INTMASK_OFFSET);
255 
256 	if (ndev->ntb.topo == NTB_TOPO_SEC)
257 		return -EINVAL;
258 	dev_dbg(ndev_dev(ndev), "Enabling Link.\n");
259 
260 	ntb_ctl = readl(mmio + AMD_CNTL_OFFSET);
261 	ntb_ctl |= (PMM_REG_CTL | SMM_REG_CTL);
262 	writel(ntb_ctl, mmio + AMD_CNTL_OFFSET);
263 
264 	return 0;
265 }
266 
267 static int amd_ntb_link_disable(struct ntb_dev *ntb)
268 {
269 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
270 	void __iomem *mmio = ndev->self_mmio;
271 	u32 ntb_ctl;
272 
273 	/* Disable event interrupt */
274 	ndev->int_mask |= AMD_EVENT_INTMASK;
275 	writel(ndev->int_mask, mmio + AMD_INTMASK_OFFSET);
276 
277 	if (ndev->ntb.topo == NTB_TOPO_SEC)
278 		return -EINVAL;
279 	dev_dbg(ndev_dev(ndev), "Enabling Link.\n");
280 
281 	ntb_ctl = readl(mmio + AMD_CNTL_OFFSET);
282 	ntb_ctl &= ~(PMM_REG_CTL | SMM_REG_CTL);
283 	writel(ntb_ctl, mmio + AMD_CNTL_OFFSET);
284 
285 	return 0;
286 }
287 
288 static u64 amd_ntb_db_valid_mask(struct ntb_dev *ntb)
289 {
290 	return ntb_ndev(ntb)->db_valid_mask;
291 }
292 
293 static int amd_ntb_db_vector_count(struct ntb_dev *ntb)
294 {
295 	return ntb_ndev(ntb)->db_count;
296 }
297 
298 static u64 amd_ntb_db_vector_mask(struct ntb_dev *ntb, int db_vector)
299 {
300 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
301 
302 	if (db_vector < 0 || db_vector > ndev->db_count)
303 		return 0;
304 
305 	return ntb_ndev(ntb)->db_valid_mask & (1 << db_vector);
306 }
307 
308 static u64 amd_ntb_db_read(struct ntb_dev *ntb)
309 {
310 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
311 	void __iomem *mmio = ndev->self_mmio;
312 
313 	return (u64)readw(mmio + AMD_DBSTAT_OFFSET);
314 }
315 
316 static int amd_ntb_db_clear(struct ntb_dev *ntb, u64 db_bits)
317 {
318 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
319 	void __iomem *mmio = ndev->self_mmio;
320 
321 	writew((u16)db_bits, mmio + AMD_DBSTAT_OFFSET);
322 
323 	return 0;
324 }
325 
326 static int amd_ntb_db_set_mask(struct ntb_dev *ntb, u64 db_bits)
327 {
328 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
329 	void __iomem *mmio = ndev->self_mmio;
330 	unsigned long flags;
331 
332 	if (db_bits & ~ndev->db_valid_mask)
333 		return -EINVAL;
334 
335 	spin_lock_irqsave(&ndev->db_mask_lock, flags);
336 	ndev->db_mask |= db_bits;
337 	writew((u16)ndev->db_mask, mmio + AMD_DBMASK_OFFSET);
338 	spin_unlock_irqrestore(&ndev->db_mask_lock, flags);
339 
340 	return 0;
341 }
342 
343 static int amd_ntb_db_clear_mask(struct ntb_dev *ntb, u64 db_bits)
344 {
345 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
346 	void __iomem *mmio = ndev->self_mmio;
347 	unsigned long flags;
348 
349 	if (db_bits & ~ndev->db_valid_mask)
350 		return -EINVAL;
351 
352 	spin_lock_irqsave(&ndev->db_mask_lock, flags);
353 	ndev->db_mask &= ~db_bits;
354 	writew((u16)ndev->db_mask, mmio + AMD_DBMASK_OFFSET);
355 	spin_unlock_irqrestore(&ndev->db_mask_lock, flags);
356 
357 	return 0;
358 }
359 
360 static int amd_ntb_peer_db_addr(struct ntb_dev *ntb,
361 				phys_addr_t *db_addr,
362 				resource_size_t *db_size)
363 {
364 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
365 
366 	if (db_addr)
367 		*db_addr = (phys_addr_t)(ndev->peer_mmio + AMD_DBREQ_OFFSET);
368 	if (db_size)
369 		*db_size = sizeof(u32);
370 
371 	return 0;
372 }
373 
374 static int amd_ntb_peer_db_set(struct ntb_dev *ntb, u64 db_bits)
375 {
376 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
377 	void __iomem *mmio = ndev->self_mmio;
378 
379 	writew((u16)db_bits, mmio + AMD_DBREQ_OFFSET);
380 
381 	return 0;
382 }
383 
384 static int amd_ntb_spad_count(struct ntb_dev *ntb)
385 {
386 	return ntb_ndev(ntb)->spad_count;
387 }
388 
389 static u32 amd_ntb_spad_read(struct ntb_dev *ntb, int idx)
390 {
391 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
392 	void __iomem *mmio = ndev->self_mmio;
393 	u32 offset;
394 
395 	if (idx < 0 || idx >= ndev->spad_count)
396 		return 0;
397 
398 	offset = ndev->self_spad + (idx << 2);
399 	return readl(mmio + AMD_SPAD_OFFSET + offset);
400 }
401 
402 static int amd_ntb_spad_write(struct ntb_dev *ntb,
403 			      int idx, u32 val)
404 {
405 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
406 	void __iomem *mmio = ndev->self_mmio;
407 	u32 offset;
408 
409 	if (idx < 0 || idx >= ndev->spad_count)
410 		return -EINVAL;
411 
412 	offset = ndev->self_spad + (idx << 2);
413 	writel(val, mmio + AMD_SPAD_OFFSET + offset);
414 
415 	return 0;
416 }
417 
418 static int amd_ntb_peer_spad_addr(struct ntb_dev *ntb, int idx,
419 				  phys_addr_t *spad_addr)
420 {
421 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
422 
423 	if (idx < 0 || idx >= ndev->spad_count)
424 		return -EINVAL;
425 
426 	if (spad_addr)
427 		*spad_addr = (phys_addr_t)(ndev->self_mmio + AMD_SPAD_OFFSET +
428 					   ndev->peer_spad + (idx << 2));
429 	return 0;
430 }
431 
432 static u32 amd_ntb_peer_spad_read(struct ntb_dev *ntb, int idx)
433 {
434 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
435 	void __iomem *mmio = ndev->self_mmio;
436 	u32 offset;
437 
438 	if (idx < 0 || idx >= ndev->spad_count)
439 		return -EINVAL;
440 
441 	offset = ndev->peer_spad + (idx << 2);
442 	return readl(mmio + AMD_SPAD_OFFSET + offset);
443 }
444 
445 static int amd_ntb_peer_spad_write(struct ntb_dev *ntb,
446 				   int idx, u32 val)
447 {
448 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
449 	void __iomem *mmio = ndev->self_mmio;
450 	u32 offset;
451 
452 	if (idx < 0 || idx >= ndev->spad_count)
453 		return -EINVAL;
454 
455 	offset = ndev->peer_spad + (idx << 2);
456 	writel(val, mmio + AMD_SPAD_OFFSET + offset);
457 
458 	return 0;
459 }
460 
461 static const struct ntb_dev_ops amd_ntb_ops = {
462 	.mw_count		= amd_ntb_mw_count,
463 	.mw_get_range		= amd_ntb_mw_get_range,
464 	.mw_set_trans		= amd_ntb_mw_set_trans,
465 	.link_is_up		= amd_ntb_link_is_up,
466 	.link_enable		= amd_ntb_link_enable,
467 	.link_disable		= amd_ntb_link_disable,
468 	.db_valid_mask		= amd_ntb_db_valid_mask,
469 	.db_vector_count	= amd_ntb_db_vector_count,
470 	.db_vector_mask		= amd_ntb_db_vector_mask,
471 	.db_read		= amd_ntb_db_read,
472 	.db_clear		= amd_ntb_db_clear,
473 	.db_set_mask		= amd_ntb_db_set_mask,
474 	.db_clear_mask		= amd_ntb_db_clear_mask,
475 	.peer_db_addr		= amd_ntb_peer_db_addr,
476 	.peer_db_set		= amd_ntb_peer_db_set,
477 	.spad_count		= amd_ntb_spad_count,
478 	.spad_read		= amd_ntb_spad_read,
479 	.spad_write		= amd_ntb_spad_write,
480 	.peer_spad_addr		= amd_ntb_peer_spad_addr,
481 	.peer_spad_read		= amd_ntb_peer_spad_read,
482 	.peer_spad_write	= amd_ntb_peer_spad_write,
483 };
484 
485 static void amd_ack_smu(struct amd_ntb_dev *ndev, u32 bit)
486 {
487 	void __iomem *mmio = ndev->self_mmio;
488 	int reg;
489 
490 	reg = readl(mmio + AMD_SMUACK_OFFSET);
491 	reg |= bit;
492 	writel(reg, mmio + AMD_SMUACK_OFFSET);
493 
494 	ndev->peer_sta |= bit;
495 }
496 
497 static void amd_handle_event(struct amd_ntb_dev *ndev, int vec)
498 {
499 	void __iomem *mmio = ndev->self_mmio;
500 	u32 status;
501 
502 	status = readl(mmio + AMD_INTSTAT_OFFSET);
503 	if (!(status & AMD_EVENT_INTMASK))
504 		return;
505 
506 	dev_dbg(ndev_dev(ndev), "status = 0x%x and vec = %d\n", status, vec);
507 
508 	status &= AMD_EVENT_INTMASK;
509 	switch (status) {
510 	case AMD_PEER_FLUSH_EVENT:
511 		dev_info(ndev_dev(ndev), "Flush is done.\n");
512 		break;
513 	case AMD_PEER_RESET_EVENT:
514 		amd_ack_smu(ndev, AMD_PEER_RESET_EVENT);
515 
516 		/* link down first */
517 		ntb_link_event(&ndev->ntb);
518 		/* polling peer status */
519 		schedule_delayed_work(&ndev->hb_timer, AMD_LINK_HB_TIMEOUT);
520 
521 		break;
522 	case AMD_PEER_D3_EVENT:
523 	case AMD_PEER_PMETO_EVENT:
524 		amd_ack_smu(ndev, status);
525 
526 		/* link down */
527 		ntb_link_event(&ndev->ntb);
528 
529 		break;
530 	case AMD_PEER_D0_EVENT:
531 		mmio = ndev->peer_mmio;
532 		status = readl(mmio + AMD_PMESTAT_OFFSET);
533 		/* check if this is WAKEUP event */
534 		if (status & 0x1)
535 			dev_info(ndev_dev(ndev), "Wakeup is done.\n");
536 
537 		amd_ack_smu(ndev, AMD_PEER_D0_EVENT);
538 
539 		/* start a timer to poll link status */
540 		schedule_delayed_work(&ndev->hb_timer,
541 				      AMD_LINK_HB_TIMEOUT);
542 		break;
543 	default:
544 		dev_info(ndev_dev(ndev), "event status = 0x%x.\n", status);
545 		break;
546 	}
547 }
548 
549 static irqreturn_t ndev_interrupt(struct amd_ntb_dev *ndev, int vec)
550 {
551 	dev_dbg(ndev_dev(ndev), "vec %d\n", vec);
552 
553 	if (vec > (AMD_DB_CNT - 1) || (ndev->msix_vec_count == 1))
554 		amd_handle_event(ndev, vec);
555 
556 	if (vec < AMD_DB_CNT)
557 		ntb_db_event(&ndev->ntb, vec);
558 
559 	return IRQ_HANDLED;
560 }
561 
562 static irqreturn_t ndev_vec_isr(int irq, void *dev)
563 {
564 	struct amd_ntb_vec *nvec = dev;
565 
566 	return ndev_interrupt(nvec->ndev, nvec->num);
567 }
568 
569 static irqreturn_t ndev_irq_isr(int irq, void *dev)
570 {
571 	struct amd_ntb_dev *ndev = dev;
572 
573 	return ndev_interrupt(ndev, irq - ndev_pdev(ndev)->irq);
574 }
575 
576 static int ndev_init_isr(struct amd_ntb_dev *ndev,
577 			 int msix_min, int msix_max)
578 {
579 	struct pci_dev *pdev;
580 	int rc, i, msix_count, node;
581 
582 	pdev = ndev_pdev(ndev);
583 
584 	node = dev_to_node(&pdev->dev);
585 
586 	ndev->db_mask = ndev->db_valid_mask;
587 
588 	/* Try to set up msix irq */
589 	ndev->vec = kzalloc_node(msix_max * sizeof(*ndev->vec),
590 				 GFP_KERNEL, node);
591 	if (!ndev->vec)
592 		goto err_msix_vec_alloc;
593 
594 	ndev->msix = kzalloc_node(msix_max * sizeof(*ndev->msix),
595 				  GFP_KERNEL, node);
596 	if (!ndev->msix)
597 		goto err_msix_alloc;
598 
599 	for (i = 0; i < msix_max; ++i)
600 		ndev->msix[i].entry = i;
601 
602 	msix_count = pci_enable_msix_range(pdev, ndev->msix,
603 					   msix_min, msix_max);
604 	if (msix_count < 0)
605 		goto err_msix_enable;
606 
607 	/* NOTE: Disable MSIX if msix count is less than 16 because of
608 	 * hardware limitation.
609 	 */
610 	if (msix_count < msix_min) {
611 		pci_disable_msix(pdev);
612 		goto err_msix_enable;
613 	}
614 
615 	for (i = 0; i < msix_count; ++i) {
616 		ndev->vec[i].ndev = ndev;
617 		ndev->vec[i].num = i;
618 		rc = request_irq(ndev->msix[i].vector, ndev_vec_isr, 0,
619 				 "ndev_vec_isr", &ndev->vec[i]);
620 		if (rc)
621 			goto err_msix_request;
622 	}
623 
624 	dev_dbg(ndev_dev(ndev), "Using msix interrupts\n");
625 	ndev->db_count = msix_min;
626 	ndev->msix_vec_count = msix_max;
627 	return 0;
628 
629 err_msix_request:
630 	while (i-- > 0)
631 		free_irq(ndev->msix[i].vector, ndev);
632 	pci_disable_msix(pdev);
633 err_msix_enable:
634 	kfree(ndev->msix);
635 err_msix_alloc:
636 	kfree(ndev->vec);
637 err_msix_vec_alloc:
638 	ndev->msix = NULL;
639 	ndev->vec = NULL;
640 
641 	/* Try to set up msi irq */
642 	rc = pci_enable_msi(pdev);
643 	if (rc)
644 		goto err_msi_enable;
645 
646 	rc = request_irq(pdev->irq, ndev_irq_isr, 0,
647 			 "ndev_irq_isr", ndev);
648 	if (rc)
649 		goto err_msi_request;
650 
651 	dev_dbg(ndev_dev(ndev), "Using msi interrupts\n");
652 	ndev->db_count = 1;
653 	ndev->msix_vec_count = 1;
654 	return 0;
655 
656 err_msi_request:
657 	pci_disable_msi(pdev);
658 err_msi_enable:
659 
660 	/* Try to set up intx irq */
661 	pci_intx(pdev, 1);
662 
663 	rc = request_irq(pdev->irq, ndev_irq_isr, IRQF_SHARED,
664 			 "ndev_irq_isr", ndev);
665 	if (rc)
666 		goto err_intx_request;
667 
668 	dev_dbg(ndev_dev(ndev), "Using intx interrupts\n");
669 	ndev->db_count = 1;
670 	ndev->msix_vec_count = 1;
671 	return 0;
672 
673 err_intx_request:
674 	return rc;
675 }
676 
677 static void ndev_deinit_isr(struct amd_ntb_dev *ndev)
678 {
679 	struct pci_dev *pdev;
680 	void __iomem *mmio = ndev->self_mmio;
681 	int i;
682 
683 	pdev = ndev_pdev(ndev);
684 
685 	/* Mask all doorbell interrupts */
686 	ndev->db_mask = ndev->db_valid_mask;
687 	writel(ndev->db_mask, mmio + AMD_DBMASK_OFFSET);
688 
689 	if (ndev->msix) {
690 		i = ndev->msix_vec_count;
691 		while (i--)
692 			free_irq(ndev->msix[i].vector, &ndev->vec[i]);
693 		pci_disable_msix(pdev);
694 		kfree(ndev->msix);
695 		kfree(ndev->vec);
696 	} else {
697 		free_irq(pdev->irq, ndev);
698 		if (pci_dev_msi_enabled(pdev))
699 			pci_disable_msi(pdev);
700 		else
701 			pci_intx(pdev, 0);
702 	}
703 }
704 
705 static ssize_t ndev_debugfs_read(struct file *filp, char __user *ubuf,
706 				 size_t count, loff_t *offp)
707 {
708 	struct amd_ntb_dev *ndev;
709 	void __iomem *mmio;
710 	char *buf;
711 	size_t buf_size;
712 	ssize_t ret, off;
713 	union { u64 v64; u32 v32; u16 v16; } u;
714 
715 	ndev = filp->private_data;
716 	mmio = ndev->self_mmio;
717 
718 	buf_size = min(count, 0x800ul);
719 
720 	buf = kmalloc(buf_size, GFP_KERNEL);
721 	if (!buf)
722 		return -ENOMEM;
723 
724 	off = 0;
725 
726 	off += scnprintf(buf + off, buf_size - off,
727 			 "NTB Device Information:\n");
728 
729 	off += scnprintf(buf + off, buf_size - off,
730 			 "Connection Topology -\t%s\n",
731 			 ntb_topo_string(ndev->ntb.topo));
732 
733 	off += scnprintf(buf + off, buf_size - off,
734 			 "LNK STA -\t\t%#06x\n", ndev->lnk_sta);
735 
736 	if (!amd_link_is_up(ndev)) {
737 		off += scnprintf(buf + off, buf_size - off,
738 				 "Link Status -\t\tDown\n");
739 	} else {
740 		off += scnprintf(buf + off, buf_size - off,
741 				 "Link Status -\t\tUp\n");
742 		off += scnprintf(buf + off, buf_size - off,
743 				 "Link Speed -\t\tPCI-E Gen %u\n",
744 				 NTB_LNK_STA_SPEED(ndev->lnk_sta));
745 		off += scnprintf(buf + off, buf_size - off,
746 				 "Link Width -\t\tx%u\n",
747 				 NTB_LNK_STA_WIDTH(ndev->lnk_sta));
748 	}
749 
750 	off += scnprintf(buf + off, buf_size - off,
751 			 "Memory Window Count -\t%u\n", ndev->mw_count);
752 	off += scnprintf(buf + off, buf_size - off,
753 			 "Scratchpad Count -\t%u\n", ndev->spad_count);
754 	off += scnprintf(buf + off, buf_size - off,
755 			 "Doorbell Count -\t%u\n", ndev->db_count);
756 	off += scnprintf(buf + off, buf_size - off,
757 			 "MSIX Vector Count -\t%u\n", ndev->msix_vec_count);
758 
759 	off += scnprintf(buf + off, buf_size - off,
760 			 "Doorbell Valid Mask -\t%#llx\n", ndev->db_valid_mask);
761 
762 	u.v32 = readl(ndev->self_mmio + AMD_DBMASK_OFFSET);
763 	off += scnprintf(buf + off, buf_size - off,
764 			 "Doorbell Mask -\t\t\t%#06x\n", u.v32);
765 
766 	u.v32 = readl(mmio + AMD_DBSTAT_OFFSET);
767 	off += scnprintf(buf + off, buf_size - off,
768 			 "Doorbell Bell -\t\t\t%#06x\n", u.v32);
769 
770 	off += scnprintf(buf + off, buf_size - off,
771 			 "\nNTB Incoming XLAT:\n");
772 
773 	u.v64 = read64(mmio + AMD_BAR1XLAT_OFFSET);
774 	off += scnprintf(buf + off, buf_size - off,
775 			 "XLAT1 -\t\t%#018llx\n", u.v64);
776 
777 	u.v64 = read64(ndev->self_mmio + AMD_BAR23XLAT_OFFSET);
778 	off += scnprintf(buf + off, buf_size - off,
779 			 "XLAT23 -\t\t%#018llx\n", u.v64);
780 
781 	u.v64 = read64(ndev->self_mmio + AMD_BAR45XLAT_OFFSET);
782 	off += scnprintf(buf + off, buf_size - off,
783 			 "XLAT45 -\t\t%#018llx\n", u.v64);
784 
785 	u.v32 = readl(mmio + AMD_BAR1LMT_OFFSET);
786 	off += scnprintf(buf + off, buf_size - off,
787 			 "LMT1 -\t\t\t%#06x\n", u.v32);
788 
789 	u.v64 = read64(ndev->self_mmio + AMD_BAR23LMT_OFFSET);
790 	off += scnprintf(buf + off, buf_size - off,
791 			 "LMT23 -\t\t\t%#018llx\n", u.v64);
792 
793 	u.v64 = read64(ndev->self_mmio + AMD_BAR45LMT_OFFSET);
794 	off += scnprintf(buf + off, buf_size - off,
795 			 "LMT45 -\t\t\t%#018llx\n", u.v64);
796 
797 	ret = simple_read_from_buffer(ubuf, count, offp, buf, off);
798 	kfree(buf);
799 	return ret;
800 }
801 
802 static void ndev_init_debugfs(struct amd_ntb_dev *ndev)
803 {
804 	if (!debugfs_dir) {
805 		ndev->debugfs_dir = NULL;
806 		ndev->debugfs_info = NULL;
807 	} else {
808 		ndev->debugfs_dir =
809 			debugfs_create_dir(ndev_name(ndev), debugfs_dir);
810 		if (!ndev->debugfs_dir)
811 			ndev->debugfs_info = NULL;
812 		else
813 			ndev->debugfs_info =
814 				debugfs_create_file("info", S_IRUSR,
815 						    ndev->debugfs_dir, ndev,
816 						    &amd_ntb_debugfs_info);
817 	}
818 }
819 
820 static void ndev_deinit_debugfs(struct amd_ntb_dev *ndev)
821 {
822 	debugfs_remove_recursive(ndev->debugfs_dir);
823 }
824 
825 static inline void ndev_init_struct(struct amd_ntb_dev *ndev,
826 				    struct pci_dev *pdev)
827 {
828 	ndev->ntb.pdev = pdev;
829 	ndev->ntb.topo = NTB_TOPO_NONE;
830 	ndev->ntb.ops = &amd_ntb_ops;
831 	ndev->int_mask = AMD_EVENT_INTMASK;
832 	spin_lock_init(&ndev->db_mask_lock);
833 }
834 
835 static int amd_poll_link(struct amd_ntb_dev *ndev)
836 {
837 	void __iomem *mmio = ndev->peer_mmio;
838 	u32 reg, stat;
839 	int rc;
840 
841 	reg = readl(mmio + AMD_SIDEINFO_OFFSET);
842 	reg &= NTB_LIN_STA_ACTIVE_BIT;
843 
844 	dev_dbg(ndev_dev(ndev), "%s: reg_val = 0x%x.\n", __func__, reg);
845 
846 	if (reg == ndev->cntl_sta)
847 		return 0;
848 
849 	ndev->cntl_sta = reg;
850 
851 	rc = pci_read_config_dword(ndev->ntb.pdev,
852 				   AMD_LINK_STATUS_OFFSET, &stat);
853 	if (rc)
854 		return 0;
855 	ndev->lnk_sta = stat;
856 
857 	return 1;
858 }
859 
860 static void amd_link_hb(struct work_struct *work)
861 {
862 	struct amd_ntb_dev *ndev = hb_ndev(work);
863 
864 	if (amd_poll_link(ndev))
865 		ntb_link_event(&ndev->ntb);
866 
867 	if (!amd_link_is_up(ndev))
868 		schedule_delayed_work(&ndev->hb_timer, AMD_LINK_HB_TIMEOUT);
869 }
870 
871 static int amd_init_isr(struct amd_ntb_dev *ndev)
872 {
873 	return ndev_init_isr(ndev, AMD_DB_CNT, AMD_MSIX_VECTOR_CNT);
874 }
875 
876 static void amd_init_side_info(struct amd_ntb_dev *ndev)
877 {
878 	void __iomem *mmio = ndev->self_mmio;
879 	unsigned int reg;
880 
881 	reg = readl(mmio + AMD_SIDEINFO_OFFSET);
882 	if (!(reg & AMD_SIDE_READY)) {
883 		reg |= AMD_SIDE_READY;
884 		writel(reg, mmio + AMD_SIDEINFO_OFFSET);
885 	}
886 }
887 
888 static void amd_deinit_side_info(struct amd_ntb_dev *ndev)
889 {
890 	void __iomem *mmio = ndev->self_mmio;
891 	unsigned int reg;
892 
893 	reg = readl(mmio + AMD_SIDEINFO_OFFSET);
894 	if (reg & AMD_SIDE_READY) {
895 		reg &= ~AMD_SIDE_READY;
896 		writel(reg, mmio + AMD_SIDEINFO_OFFSET);
897 		readl(mmio + AMD_SIDEINFO_OFFSET);
898 	}
899 }
900 
901 static int amd_init_ntb(struct amd_ntb_dev *ndev)
902 {
903 	void __iomem *mmio = ndev->self_mmio;
904 
905 	ndev->mw_count = AMD_MW_CNT;
906 	ndev->spad_count = AMD_SPADS_CNT;
907 	ndev->db_count = AMD_DB_CNT;
908 
909 	switch (ndev->ntb.topo) {
910 	case NTB_TOPO_PRI:
911 	case NTB_TOPO_SEC:
912 		ndev->spad_count >>= 1;
913 		if (ndev->ntb.topo == NTB_TOPO_PRI) {
914 			ndev->self_spad = 0;
915 			ndev->peer_spad = 0x20;
916 		} else {
917 			ndev->self_spad = 0x20;
918 			ndev->peer_spad = 0;
919 		}
920 
921 		INIT_DELAYED_WORK(&ndev->hb_timer, amd_link_hb);
922 		schedule_delayed_work(&ndev->hb_timer, AMD_LINK_HB_TIMEOUT);
923 
924 		break;
925 	default:
926 		dev_err(ndev_dev(ndev), "AMD NTB does not support B2B mode.\n");
927 		return -EINVAL;
928 	}
929 
930 	ndev->db_valid_mask = BIT_ULL(ndev->db_count) - 1;
931 
932 	/* Mask event interrupts */
933 	writel(ndev->int_mask, mmio + AMD_INTMASK_OFFSET);
934 
935 	return 0;
936 }
937 
938 static enum ntb_topo amd_get_topo(struct amd_ntb_dev *ndev)
939 {
940 	void __iomem *mmio = ndev->self_mmio;
941 	u32 info;
942 
943 	info = readl(mmio + AMD_SIDEINFO_OFFSET);
944 	if (info & AMD_SIDE_MASK)
945 		return NTB_TOPO_SEC;
946 	else
947 		return NTB_TOPO_PRI;
948 }
949 
950 static int amd_init_dev(struct amd_ntb_dev *ndev)
951 {
952 	struct pci_dev *pdev;
953 	int rc = 0;
954 
955 	pdev = ndev_pdev(ndev);
956 
957 	ndev->ntb.topo = amd_get_topo(ndev);
958 	dev_dbg(ndev_dev(ndev), "AMD NTB topo is %s\n",
959 		ntb_topo_string(ndev->ntb.topo));
960 
961 	rc = amd_init_ntb(ndev);
962 	if (rc)
963 		return rc;
964 
965 	rc = amd_init_isr(ndev);
966 	if (rc) {
967 		dev_err(ndev_dev(ndev), "fail to init isr.\n");
968 		return rc;
969 	}
970 
971 	ndev->db_valid_mask = BIT_ULL(ndev->db_count) - 1;
972 
973 	return 0;
974 }
975 
976 static void amd_deinit_dev(struct amd_ntb_dev *ndev)
977 {
978 	cancel_delayed_work_sync(&ndev->hb_timer);
979 
980 	ndev_deinit_isr(ndev);
981 }
982 
983 static int amd_ntb_init_pci(struct amd_ntb_dev *ndev,
984 			    struct pci_dev *pdev)
985 {
986 	int rc;
987 
988 	pci_set_drvdata(pdev, ndev);
989 
990 	rc = pci_enable_device(pdev);
991 	if (rc)
992 		goto err_pci_enable;
993 
994 	rc = pci_request_regions(pdev, NTB_NAME);
995 	if (rc)
996 		goto err_pci_regions;
997 
998 	pci_set_master(pdev);
999 
1000 	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1001 	if (rc) {
1002 		rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1003 		if (rc)
1004 			goto err_dma_mask;
1005 		dev_warn(ndev_dev(ndev), "Cannot DMA highmem\n");
1006 	}
1007 
1008 	rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1009 	if (rc) {
1010 		rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1011 		if (rc)
1012 			goto err_dma_mask;
1013 		dev_warn(ndev_dev(ndev), "Cannot DMA consistent highmem\n");
1014 	}
1015 
1016 	ndev->self_mmio = pci_iomap(pdev, 0, 0);
1017 	if (!ndev->self_mmio) {
1018 		rc = -EIO;
1019 		goto err_dma_mask;
1020 	}
1021 	ndev->peer_mmio = ndev->self_mmio + AMD_PEER_OFFSET;
1022 
1023 	return 0;
1024 
1025 err_dma_mask:
1026 	pci_clear_master(pdev);
1027 err_pci_regions:
1028 	pci_disable_device(pdev);
1029 err_pci_enable:
1030 	pci_set_drvdata(pdev, NULL);
1031 	return rc;
1032 }
1033 
1034 static void amd_ntb_deinit_pci(struct amd_ntb_dev *ndev)
1035 {
1036 	struct pci_dev *pdev = ndev_pdev(ndev);
1037 
1038 	pci_iounmap(pdev, ndev->self_mmio);
1039 
1040 	pci_clear_master(pdev);
1041 	pci_release_regions(pdev);
1042 	pci_disable_device(pdev);
1043 	pci_set_drvdata(pdev, NULL);
1044 }
1045 
1046 static int amd_ntb_pci_probe(struct pci_dev *pdev,
1047 			     const struct pci_device_id *id)
1048 {
1049 	struct amd_ntb_dev *ndev;
1050 	int rc, node;
1051 
1052 	node = dev_to_node(&pdev->dev);
1053 
1054 	ndev = kzalloc_node(sizeof(*ndev), GFP_KERNEL, node);
1055 	if (!ndev) {
1056 		rc = -ENOMEM;
1057 		goto err_ndev;
1058 	}
1059 
1060 	ndev_init_struct(ndev, pdev);
1061 
1062 	rc = amd_ntb_init_pci(ndev, pdev);
1063 	if (rc)
1064 		goto err_init_pci;
1065 
1066 	rc = amd_init_dev(ndev);
1067 	if (rc)
1068 		goto err_init_dev;
1069 
1070 	/* write side info */
1071 	amd_init_side_info(ndev);
1072 
1073 	amd_poll_link(ndev);
1074 
1075 	ndev_init_debugfs(ndev);
1076 
1077 	rc = ntb_register_device(&ndev->ntb);
1078 	if (rc)
1079 		goto err_register;
1080 
1081 	dev_info(&pdev->dev, "NTB device registered.\n");
1082 
1083 	return 0;
1084 
1085 err_register:
1086 	ndev_deinit_debugfs(ndev);
1087 	amd_deinit_dev(ndev);
1088 err_init_dev:
1089 	amd_ntb_deinit_pci(ndev);
1090 err_init_pci:
1091 	kfree(ndev);
1092 err_ndev:
1093 	return rc;
1094 }
1095 
1096 static void amd_ntb_pci_remove(struct pci_dev *pdev)
1097 {
1098 	struct amd_ntb_dev *ndev = pci_get_drvdata(pdev);
1099 
1100 	ntb_unregister_device(&ndev->ntb);
1101 	ndev_deinit_debugfs(ndev);
1102 	amd_deinit_side_info(ndev);
1103 	amd_deinit_dev(ndev);
1104 	amd_ntb_deinit_pci(ndev);
1105 	kfree(ndev);
1106 }
1107 
1108 static const struct file_operations amd_ntb_debugfs_info = {
1109 	.owner = THIS_MODULE,
1110 	.open = simple_open,
1111 	.read = ndev_debugfs_read,
1112 };
1113 
1114 static const struct pci_device_id amd_ntb_pci_tbl[] = {
1115 	{PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_NTB)},
1116 	{0}
1117 };
1118 MODULE_DEVICE_TABLE(pci, amd_ntb_pci_tbl);
1119 
1120 static struct pci_driver amd_ntb_pci_driver = {
1121 	.name		= KBUILD_MODNAME,
1122 	.id_table	= amd_ntb_pci_tbl,
1123 	.probe		= amd_ntb_pci_probe,
1124 	.remove		= amd_ntb_pci_remove,
1125 };
1126 
1127 static int __init amd_ntb_pci_driver_init(void)
1128 {
1129 	pr_info("%s %s\n", NTB_DESC, NTB_VER);
1130 
1131 	if (debugfs_initialized())
1132 		debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
1133 
1134 	return pci_register_driver(&amd_ntb_pci_driver);
1135 }
1136 module_init(amd_ntb_pci_driver_init);
1137 
1138 static void __exit amd_ntb_pci_driver_exit(void)
1139 {
1140 	pci_unregister_driver(&amd_ntb_pci_driver);
1141 	debugfs_remove_recursive(debugfs_dir);
1142 }
1143 module_exit(amd_ntb_pci_driver_exit);
1144